Skip to content
by·step
Get started free

PLAN · BUILD · REVIEW · PUBLISH

New: agent teams, AI choices, and Desktop

From idea to app. One step at a time.

Build with your teammates and their own AI, or orchestrate agent teams in the cloud or on your computer. Share plans, tasks, evidence and reviews in one workspace.

Start freeSee how it works
  • Connected plans & tasks
  • Cloud or your computer
  • Results you can review
  • Free to get started
Bystep roadmap: a PRD on the left, feature canvas in the middle, Bystep Agent on the right

Works with the agent you already use

  • Claude Code
  • Cursor
  • Codex
  • GitHub Copilot
  • Gemini CLI
  • OpenCode
  • Antigravity
  • Pi

Connect agents through CLI or MCP. Team execution with a subscription login currently supports Claude Code and Codex on your computer.

THE WORKFLOW

From plan to progress. One workspace.

Give your idea a clear direction.

Answer questions, add references, and review your requirements before building.

Start free
Give your idea a clear direction.
Screenshots of Bystep. Choose a view to explore the workflow.

Give your idea a clear direction.

Scroll the image to explore workspace details.

Give your idea a clear direction.

YOUR WAY TO BUILD

One plan. Three ways to bring it to life.

Choose what fits your experience. Your plan, progress, and check results stay together in Bystep.

01

Automatic

Bystep builds step by step. Preview your app and review the results in your browser.

For projects supported by automatic mode. Usage depends on your allowance and AI connection.Explore automatic building
02

Guided

Need a second opinion? Get a review, polishing, or a working session to move forward.

A separate service from your subscription. Choose the support and review its cost before booking.Explore guided support
03

With your agent

Connect your computer through Desktop or CLI and work through the plan with your chosen agent. Bystep prepares the tasks, context, and checks.

Use the official agent login or your own API key. CLI and MCP integrations remain available for your development tools.Connect your computer

Team Build · for teams of 2–10

Your team. Your own AI. One view of the work.

Build with your cofounder and developers. Give each task a human owner, let everyone use their own coding tool, and follow the evidence, blockers and reviews in Bystep.

Open your team workspaceSee the team setup guide

Bring a PRD

Review source requirements and coverage before creating tasks.

Start with a brief

Agree on the audience and first outcome; keep open questions visible.

Continue an existing codebase

Sync its manifest and review proposed changes alongside existing work.

Make the next step clear for everyone

  1. Assign a person and a reviewer

    A claim reserves the task for one developer. Renew it while working or hand it off with context.

  2. Work with your own AI

    Fetch current context through the CLI or MCP, then use the coding tool you already know.

  3. Submit evidence for review

    An independent reviewer checks the current criteria and revision. Changed scope needs fresh review.

  4. Follow the whole project

    See milestones, blockers, review queues and updates that need your attention.

Illustrative team board · example work, not live project data

Milestone: first shared release

  • Build the invitation flow

    In progress

    Owner: frontend developer

    Personal AI · active claim

  • Check project permissions

    In review

    Owner: backend developer

    Test evidence submitted · teammate reviewing

  • Agree on the first milestone

    Blocked

    Owner: founder

    Product decision needed before the next task

Passing tests, human acceptance and a recorded release are separate states. Recording a release does not deploy the project.

Keep personal accounts personal

Each developer signs in to their AI locally. A workspace invitation does not share AI credentials, billing access or another person's computer.

Reserve a budget deliberately

Choose a payer and reserve an amount per session. Unreported costs stay unknown; reservations remain until reconciled. Provider usage outside Bystep follows that account's limits.

Connect delivery evidence

Link GitHub PRs and required checks. Import and refresh GitLab.com issues with read-only access, reviewing conflicts before applying changes.

Team Build coordinates people using their own tools. Agent teams provide a separate workflow for orchestrating AI workers. Explore agent teams

AGENT TEAMS · CLOUD & YOUR COMPUTER

Choose your AI. Run your agent team.

Choose where coding runs and which AI it uses on the Build page. Bystep assigns work from the same plan and checks the combined results.

Use Bystep AI

Run coding in the cloud with your Bystep plan. Start and monitor work directly from your browser.

Step and infrastructure allowances follow your plan.

Bring your own API key

Add your AI connection for coding in the cloud or on your computer. Your selected provider bills AI usage.

The payer does not switch automatically when a provider fails. Infrastructure allowances still apply.

Use your agent subscription

Connect your computer and sign in through official Claude Code or Codex. Coding runs on your computer with that agent account.

Requires an installed agent and an account with access. Keep your computer awake.

Planning, chat, and review use a separately selected AI source: Bystep AI or your API connection. A coding agent subscription does not automatically provide API access for those stages. See the AI provider policy

AGENT ORCHESTRATION

Divide the work. Check the results together.

Select “Start agent team” when your plan is ready. Independent tasks can run together; dependent work follows its required order.

  • Set team size and usage thresholds

    Choose parallelism, token/cost thresholds, and a deadline. Work already running may exceed the usage thresholds.

  • Guide the team as work progresses

    Your messages join subsequent assignments alongside each task’s context and acceptance criteria.

  • Stop and continue with clear ownership

    Approve checkpoints to continue. When stopped, Bystep waits for confirmation that processes have ended before releasing the work.

Open a project and start a team
Example workflow for two independent tasks

  1. 01 · Assign tasks to separate working folders

    Agent 1 · Page interface
    Agent 2 · Data service

  2. 02 · Combine and check

    Results are checked together against the plan’s criteria. A process stopping does not make a task complete.

  3. 03 · Review and approve the phase

    Review the evidence, request fixes if needed, and continue according to your approval checkpoints.

YOU STAY IN CONTROL

Agents do the work. You keep the decisions.

Every control below is optional and switched on per project. Without them, the agent team runs as usual.

  • A read-only Reviewer

    A second agent reads the merged result without changing a single file, then records findings by severity. Serious findings hold the merge until they are fixed. Review allowance follows your plan.

  • Command approvals

    On your computer, agents can be told to ask before risky commands or before every command. If nobody answers in time, the command is declined.

  • Decision Cards

    When there is a choice of direction, such as a merge conflict or a Reviewer finding, you see the options and what each one leads to. Your choice is recorded and used by agents on the next task.

  • Any agent

    One agent list for every path: Claude Code, Codex, Cursor, Gemini CLI, opencode, or your own project script on your computer, each according to its capabilities. The plan and the evidence stay the same, whoever does the work.

  • An honest provider policy

    Bystep only uses paths the AI providers allow. Your subscription login stays in the official agent on your computer, not on our servers.

    See the status of each provider

HOW IT WORKS

From idea to handover in four steps

Start with your needs, agree on the steps, then build and review the results with Bystep.

STEP 1

Plan

  • Describe your idea, pick a starter kit (QRIS POS, WhatsApp OTP, Supabase RLS, booking, capstone), or connect a codebase
  • Review your requirements before building

STEP 2

Roadmap

  • Features become sub-features and phased tasks, each with pass/fail criteria
  • Choose when results need your approval

STEP 3

Build & prove

  • Choose cloud or your computer, then the AI source for coding
  • Start an agent team for independent tasks that can run together
  • Review combined results and check evidence before continuing

STEP 4

Publish & keep improving

  • Review previews and readiness for automatic projects
  • Hand over to clients with Studio, according to your plan

BYSTEP DESKTOP · OPTIONAL

Your computer, connected to Bystep.

Connect your account, sign in to an agent, and monitor the local connection in one window. Manage plans, reviews, and approvals in your browser.

Download Desktop for Linux x64Computer connection guide

.tgz archive. The latest version is in the download list. Requires Node.js 22+ and separately installed Claude Code or Codex.

The currently available and tested package is Linux x64. Windows and macOS Desktop packages are not yet available as supported releases. Cloud execution remains available through your browser.

Package versions and checksums

Three steps to connect

  1. Extract and connect your account

    Open bystep-desktop from the extracted download folder. Select “Connect account”, then confirm in your browser.

  2. Sign in to your chosen agent

    Under “Your agents”, select Sign in. Follow the provider’s official process with your own account.

  3. Start the connection, then open Build

    Select “Start connection”. Choose this computer on the Bystep web Build page, then start your work.

Keep Desktop open and your computer awake during local work. You can also connect through the CLI without Desktop.

Design Room

Agree. Plan. Prove. Ship.

A structured discovery room for teams, agency clients or capstone advisors — six async steps that turn a group chat into a PRD with a traceable origin for every feature.

See Design RoomStart free

Guests join free, no account

A WhatsApp link is enough — a name, no sign-up. Only the room owner spends quota; everyone else writes cards and votes for free.

AI challenges, it doesn't invent

Idea cards stay locked until everyone reveals together. The facilitator summarizes, challenges and critiques from a persona — it never writes a card before people do.

Every feature carries its origin

Each feature in the resulting PRD links back to the card, the hypothesis and the vote that produced it — Proof-of-Why, not just Proof-of-Done.

Explore a six-step Design Room example

Product workflow illustration

  1. 01Problem
  2. 02Users
  3. 03Assumptions
  4. 04Ideas
  5. 05Choose
  6. 06Formulate

Simulation · a salon booking room

  • DP 30% charged automatically via virtual account
  • WhatsApp reminder 1 day before the appointment
  • Book by chair, not by hour block
  • Hidden cardwritten, not yet revealed
AI Facilitator · locked until revealSummarizeChallengePersona critique
  • Virtual brainwriting beats video brainstorming — more ideas, more original, deeper elaboration (Baruah et al. 2025).
  • Ambiguous requirements cut coding-agent Pass@1 by 35–52% (Wu et al. 2024, HumanEvalComm).
  • A hypothesis-then-test approach across 759 startups (4 RCTs) produced better performance and fewer false positives (Camuffo et al. 2020/2024).

PROOF OF DONE

Proof, not claims

  • Every task carries pass/fail criteria, written into the PRD before a line of code exists.
  • bystep task verify runs them on your machine: tests, build, file checks, screenshots.
  • The server refuses task complete without passing evidence; every --override is recorded with a reason. The verifier can be Local, the Bystep Reviewer, or GitHub CI.

Product workflow illustration

BYSTEP AGENT

AI that follows your plan.

Start freeCLI reference

Your agent understands project context, proposes changes for review, and works through tasks with check evidence. You choose the approval points.

Answers from your whole plan

Ask what is left in phase 2, which tasks touch auth, or why a feature exists. It reads the plan, not a chat log.

Proposes, you approve

Every edit arrives as a proposal with a diff — including drift proposals when the code no longer matches the PRD. Apply, reject, or ask for another version.

Knows your codebase

Run bystep sync and the wiki is rebuilt from your repo, so the agent plans against the code you actually have.

Explore the technical workflow, modes, and checks
  1. 01
    bystep task next

    A packet, not a prompt

    The agent gets a Task Packet of ~1–2k tokens: the PRD excerpt, pass criteria, files it may and may not touch, conventions from the wiki.

  2. 02
    bystep task start

    Starting point recorded

    The starting commit is saved in .bystep/. Diff, duration and cost are measured from here.

  3. 03
    agent works

    Code stays on your machine

    In the local workflow, your coding agent works on your project locally. Automatic mode uses Bystep’s server environment.

  4. 04
    bystep task verify

    Criteria are executed

    test · build · lint · http · file · screenshot. Output is scrubbed of secrets, then uploaded as evidence.

  5. 05
    bystep task complete

    The server gate

    No passing evidence → refused (409). Two failures → the task tier escalates and a stronger model is suggested to the agent.

  6. 06
    task approve · wave next

    Checkpoint & waves

    You approve the phase, or return tasks to the agent with a note. Then up to 4 tasks with non-overlapping files run in parallel.

Three execution modes

Autopilot

The agent moves to the next task as long as evidence passes. For Utility tasks and small projects.

Checkpoint

Stops at the end of every phase. You approve, or send tasks back with a note. The default.

Strict

Every task waits for your approval. For auth, payments and anything touching customer data.

Three verifiers

  • Local — criteria run on your machine through the CLI
  • Bystep Reviewer — an AI reviewer reads the diff and the evidence
  • CI — a GitHub Actions webhook becomes the official evidence

Guard hook for Claude Code

bystep init installs task guard as a Stop hook: the session cannot end while a started task has no passing evidence.

Wave mode

The server picks ≤4 tasks whose files do not overlap; one leader prompt runs parallel subagents. New projects: the first wave stays sequential.

CONTROL TOWER

See where your AI budget goes — before the bill arrives

See agents, models, duration, attempts, and provider-reported tokens and costs. Bystep summarizes progress per feature, detects code drifting from the PRD, and assesses repository readiness for agents.

Cost & duration per feature

Review available usage by feature, task, and agent. Cost figures depend on the data each provider reports.

Bystep Cost & duration page: summary tiles, per-feature table, daily and per-agent charts

PRD ↔ code drift

Spec coverage is computed every time the wiki is rebuilt; a cov badge appears on roadmap nodes. When code drifts, the agent proposes a PRD revision — you apply or reject.

Agent-readiness score

Paste a public repo URL: 11 signals, 100 points, plus leaked secrets. Offline: bystep doctor --repo with the identical score.

Check a repo for free

Public benchmark

Anonymous pass rate per agent, model and task kind across all Bystep verifications. One-click opt-out in Settings.

See the benchmark

PLATFORM CAPABILITIES

All your work, connected.

From the first decision to ongoing maintenance. Open the guides to see how each capability works.

Human team collaboration

Share scope, assign owners, claim tasks, hand off context and review progress together.

Learn more

Needs your attention

One inbox for approvals, comments, and work needing help, with a clear next action.

Learn more

Rules & fixes

Set project rules. Repeated failures create a fix task with the evidence of what went wrong.

Learn more

Teams & notifications

Owners, comments, activity, and notifications with daily digests based on your settings.

Learn more

Your choice of AI

Choose a model within your plan or connect your own AI service. Provider usage is billed separately.

Learn more

Templates & existing projects

Start with local business templates or continue from an existing project and documents.

Learn more

Import, export & integrate

Move plans between tools, export documents, and connect agents through CLI, MCP, or API.

Learn more

Campus & facilitators

Design Rooms, capstone projects, participation, and rubrics support collaborative learning.

Learn more

Tools & support

Explore AI tools, comparisons, coaching, and support for your project needs.

Learn more

BYSTEP STUDIO

Evidence you can bill a client for

Your agent has already proven every task. Studio turns that evidence into what clients pay for: estimates, proposals, a progress page and a handover package.

  • Estimates from real data

    Median verified durations across projects, times your rate and a buffer — not a guess.

  • A PDF proposal in your brand

    Bilingual, versioned, with your agency’s logo and colours. Built from the same PRD and estimate.

  • A client page without an account

    Evidence bar per feature, exceptions, an Approve-phase button. Approvals are recorded in the client’s name.

  • A handover ZIP

    PRD, wiki, evidence log, overrides, estimate vs actual, deploy and security notes — one file.

Estimates · proposals · client portal · handover

Join the Studio waitlistSee pricing
Bystep client page: progress per feature with an evidence bar and an Approve-phase button

FROM PREVIEW TO APP

See it working. Publish when ready.

For automatic projects, previews, readiness checks, and release versions are part of one workflow.

Explore the full workflow

Publishing depends on project readiness and hosting limits. Local projects use your own hosting provider.

  1. 01

    Preview & check

    Try your app in the browser and review the checklist before publishing.

  2. 02

    Publish & monitor

    Publish to an app address provided by Bystep. See its status and release history.

  3. 03

    Restore & take it with you

    Restore a previous version, pause or wake your app, and download the project as a ZIP.

Product workflow illustration

CLI & AGENT-NATIVE

Your agent already knows how to use Bystep

Read the CLI docsComputer connection guide

Connect your computer and agent through CLI, skills, hooks, or MCP. Run a single task or coordinate an agent team from the same plan.

Product workflow illustration

The @bystep/cli CLI

Install the CLI from the computer guide. Connect projects, start or monitor a team, send guidance, stop, and resume workflows from your terminal.

One skill + hook, eight agents

bystep init prepares instructions and checks for supported agents, following your project rules.

An MCP server for your agent

Existing plan and task tools remain available. Six workflow controls add listing, details, start, stop, resume, and messages for your agent team.

CLI & MCP · command contracts available in the guides

YOUR CODE STAYS YOURS

Control your code and how you work.

In local mode, sync metadata and summaries you review. In automatic mode, code is built on Bystep servers for previews and publishing. AI processing follows the connection you choose.

Transparent local sync

In the CLI workflow, sync uploads the manifest and agent summaries. Review the contents before sending.

How sync works

Personal tokens you can revoke

bs_ tokens live in ~/.config/bystep with mode 0600. Revoke any of them from Settings in one click.

Manage tokens

A transparent .bystep/ folder

config.json holds ids only and is safe to commit; pending, summaries and state are git-ignored automatically.

Read the docs

Overrides & audit on record

Skipping evidence needs a written reason; it shows on the task card, the public page and the audit export. No silent “done”.

Read about Proof-of-Done

PRICING

Start free. Grow at your own pace.

Prices and allowances below come from Bystep’s active plan catalog. Choose your currency and review the details before subscribing.

Loading current plans…

See all billing periods, currencies, models, and allowances on the pricing page. Guided support and your own AI provider usage are charged separately.

See all plans

Already using another tool? Compare first.

Bystep does not replace your spec — it becomes its evidence gate. Plans come in and go out as your tool’s native files.

FAQ

Before your first step.

Answers to help you choose how to work.

Can each developer use a different AI tool?

Yes. Each developer uses their own local AI account and reports through the Bystep CLI or MCP. Claim a task, fetch current context, do the work, verify and submit for review. Claims coordinate ownership; they do not launch an AI or share subscriptions. Agent teams are a separate execution workflow.

Do we need a finished PRD to start as a team?

You can start from a PRD, a short brief or a synced codebase. Review proposed requirements, source coverage and conflicts before applying the scope. An observed codebase is not automatically treated as verified work.

Does an accepted task automatically deploy?

No. Evidence, human acceptance and release are tracked separately. The owner can record a deployment that has already completed. The record stores its revision and destination; it does not run deployment. Tasks with only manual criteria can receive independent human review without being labeled test-verified.

Which issue tracker works with Team Build today?

GitLab.com supports read-only import and refresh with a read_api token. Preview changes and resolve conflicts before applying them. Closing a GitLab issue does not accept or release a Bystep task. Linear and Jira integrations are not available in this release.

Do I need to know how to code?

You can start without coding. Describe your needs on the getting-started page, review the plan, and choose Automatic or Guided. Local mode is for people using their own coding agent and development tools.

What is included for free?

Free includes an initial allowance for planning and agent chat. See the pricing section for current limits. Automatic building, your own AI service, guided support, and hosting follow their own terms; free to start does not mean every part of the process is free.

Does my code go to Bystep servers?

In the local CLI workflow, sync sends metadata and agent-prepared summaries. In Automatic mode, project code is built and stored in Bystep’s server environment for preview and publishing. Context sent to AI providers depends on your chosen workflow and connection.

Can I continue an existing project?

Yes. Use the existing-project flow, connect a workspace through the CLI, or import a supported plan format. Automatic mode has its own stack requirements; not every codebase can run automatically.

How does Bystep decide a task is done?

Tasks have acceptance criteria and check evidence. Your execution mode controls when approval is required. Overriding a check requires a recorded reason. Evidence helps you review results; test the app for your needs before release.

Which agents and AI services can I use?

Choose Bystep AI, a supported provider API key, or the official Claude Code/Codex login for coding on your computer. Select planning and review AI separately. CLI/MCP integrations remain available for other agents according to their capabilities. Provider charges and infrastructure allowances follow your choices.

Can I work with a team or clients?

Yes. Workspaces support members, owners, comments, and activity. Studio adds estimates, proposals, a client portal, and handover. Campus supports capstones and facilitation. Access and limits depend on your plan.

What happens after publishing?

For supported automatic projects, you can see release status, restore a previous version, pause or wake the app, and export a ZIP. Review hosting limits and your project checklist before using it for production.

Can I use my existing Claude Code or Codex subscription?

Yes, through the official agent login on your computer, provided your account includes access. Connect through Desktop or CLI, then select the agent and its login on the Build page. Planning, chat, and review use Bystep AI or an API connection selected separately.

Do I need Desktop? Are Windows and macOS versions available?

Desktop is optional. Run coding in the cloud from your browser or connect a supported computer through the CLI. The current Desktop package is for Linux x64 and requires Node.js 22+ and a separately installed official agent. Windows and macOS Desktop packages are not yet available as supported releases.

How do agents work together without overwriting each other’s code?

Bystep assigns tasks by dependencies and file scope. Independent tasks use separate working folders; dependent or conflicting work runs in sequence. Results are then combined and checked before completion. You can send guidance, stop the team, and approve checkpoints to continue.

Are token and cost limits always exact?

Usage figures depend on provider reports. Token and cost thresholds stop subsequent assignments; requests already running may exceed them. Some agent subscriptions do not report monetary costs, so that information is shown as unknown. You can also set team size and a deadline.

Your next step starts here

Start freeTalk to us